Darvirian Knowledge Graph and Search Engine

• 

  Stay safe, keep your distance!

• 

  Workflow with Active Learning

• 

  KG Search Tool Prototype

• 

  Sample Knowledge Graph visualisation

• 

  Sample Knowledge Graph visualisation

• 

  Sample Word cloud visualisation

• 

  Acknowledgements

• 

  Thanks!!!

Working on a good cause!

The Darvirian team with their name and country of residence

• Henry Bol MSc (Netherlands)
• Joseph Ambrose Pagaran PhD (Austria)
• Yao Yao PhD (Belgium)
• Eva Žerovnik PhD (Slovenia)
• Reshmi Mukherjee PhD (Netherlands)
• Aurore Legendre MSc (Japan)

Chosen Challenge:

Health & Life, Research. Under this challenge, we aim at addressing issues such as:

• mapping of covid literature with perspectives of tests/medication/vaccination development
• mapping of existing approaches (meta-narratives, for ex: support via vaccine, via prevention, via alternative medicines, etc.).
  

Overview

Challenge: mapping of covid-19 literature with perspectives of tests/medication/vaccination development.

Goal: create a Knowlegde Graph Search (KGS) tool which provides researchers fast and highly relevant insights to a research question from a big data database of documents (papers, articles, review papers, etc.): "finding the needle in the haystack".

Prototype: Experts of domain knowledge provide selected keywords of a their own specialisation such as:

1. Sample Domain Expertise: the natural and chemical compounds to treat specific Covid-19 stages of disease (courtesy of Eva Žerovnik PhD)
2. Sample Domain Expertise: biotherapy for patients undergoing recovery (courtesy of Aurore Legendre MSc)
3. Sample Domain Expertise: Covid-19 drug discovery (courtesy of Reshmi Mukherjee PhD).

These keywords are taken as input to a Search & Rank Engine written in Python with pre-processed 30,000 Covid-19 research papers.

The search results has ranking of sentences, whose ranking are evaluated by the domain experts wheher they are giving high quality of few returned research papers. Both the keywords and search engine results are later fed to create a knowlege graph. The whole process is called Active Learning.

Achievements during the weekend

1. Set up a search prototype based on about 30.000 Covid papers
2. Asked 3 domain knowledge experts to evaluate model with their own search query
3. Refined the model
4. Answered the Research challenge (and found the needle)
5. And (as usual) wanted to achieve more, struggled, had fun and all for a good cause!

Business

• Market: organisations with a need to extract knowledge out of huge amount of data Search as a Service (SaaS)
• Implementation costs for setting up the solution including active learning process
• Monthly subscription including license fee for use of engine and knowledge graph
• Needed: online version of the solution (cloud solution).

Next Steps

As the prototype involves manual input of keywords and an iteration of evaluation of returned results, we plan to automate the steps from the input of keywords to the creation of Knowledge Graph. Also planned is the writing of a preprint to describe the algorithm and also present selected user cases.

We will continue to improve the Search Engine Tool and Knowledge Graph generation:

1. Create online version: easier to evaluate and use
2. Evaluate and learn from more domain experts and decision makers
3. Refine model
4. Build cases to show the great value of the solution
5. Get first happy customer!
   

Appendices

Keywords courtesy of Eva Žerovnik PhD:

• keywords = “what is correlation between SARS-CoV-2 and Alzheimer’s?”
• keywords = “virus removal by autophagy”
• keywords = “virus factories”.
• keywords = “stress granules and SARS-CoV-2"
• keywords = “membrane channels of SARS-CoV-2 Eprotein”
• keywords = “natural compounds as broad anti-virals applied to SARS-CoV-2"
• keywords = “known drugs useful for SARS-CoV-2 treatment”
• keywords = “stem cells in cancer and SARS-CoV-2"

Keywords courtesy of Reshmi Mukherjee PhD:

• keywords = “What is the binding efficacy between SARS-CoV-2 spike protein and ACE2?”
• keywords = “what is the structure of SARS-CoV2 spike protein?”
• keywords = “How do antiviral drugs work on SARS-CoV2?”
• keywords = “How do fusion inhibitors work on SARS-CoV2?”
• keywords = “What is the mechanism of action of fusion inhibitors of SARS-CoV-2-?”
• keywords = “What are the approaches to design RNA, DNA , peptide-based vaccine for SARS-CoV-2"
• keywords = “Do fatty acids like AA work as antiviral and can work on SARS-CoV-2?”
• keywords = “What are the design principles for drug of SARS-CoV-2?”

Keywords courtesy of Aurore Legendre MSc:

• keywords = “What are the differences between pneumonia, ARDS (Acute Respiratory Distress Syndrome) and SARS (Severe Acute Respiratory Syndrome) ?”
• keywords = “How coronavirus can provoke ARDS (Acute Respiratory Distress Syndrome) ?”
• keywords = “Chloroquine, Nucleotide inhibitors, lopinavir/ritonavir, etc. : drug repurposing can be the solution against the novel coronavirus ?”

Built With

• apple
• macos
• python
• unix
• windows